The effect of random seeds on results ?

[shenzgang]

When I use Titan tests I get different and very different estimates each time. Does random seeding have that much of an impact?

[tovbinm]

Yes, indeed. In order to get a predictable behavior you can set random seed in your tests. Depending on your tests structure where you set the seed might vary. For example - https://github.com/salesforce/TransmogrifAI/blob/master/helloworld/src/main/scala/com/salesforce/hw/titanic/OpTitanic.scala#L50,

[shenzgang]

Thanks for your reply! There is also the question of how to use the generated model to predict unlabeled test sample data. Are there any examples of using model prediction?

[tovbinm]

You can save a trained model, then load it later, set a new scoring reader / a new input dataset, and finally compute scores by invoking score().

You can also use transmogrifai-local for on-line serving of your model (e.g over HTTP API)

[shenzgang]

The data set used for model training is labeled column, while the test data is not labeled column. When calling score(), an exception will be thrown

[leahmcguire]

You will need to create an empty label column